Brief summary: Dawkins demonstrates a classic "Prisoner Dilemma AI tournament". No big surprise to us today, but at the time the revelation that Tit for Tat is one of -- if not *the* -- most effective strateg(y|ies) was a surprising result. He goes on to demonstrate animals employing the Tit for Tat strategy. Assumptions of generosity, with vengefulness, appear to be strongly selected for.
(Bug report: I was sent to this post via this link, and I see MAIN bolded above the title instead of DISCUSSION. The URL is misleading too, shouldn't urls of discussion posts contain "/r/discussion/" instead of "/lw"?)
(EDIT: Grognor just told me that "every discussion post has a main-style URL that bolds MAIN")
The description of the various expert programs was eerily similar to the programs submitted to the LW competition.
There's not really much variation or innovation really available in PD trials.
Isn't it just strategy stealing? Calling it tit-for-tat maybe focuses away from the fundamental reason why it wins.
It was really surprising? That's... odd. What were they expecting?
Dawkins himself said it; they believed that "sneaky cheaters" would prosper more. This is a common intuition even today. It turns out that Tit for Tat remains a very robust strategy, despite its apparent lack of sophistication, in prisoner-dilemma trials. There was a run just a month or two back that was all the rage on Discussion on LW.
It's a very robust strategy because of its apparent lack of sophistication. By being easy to detect and predict, it causes its opponents to chose cooperation. Since you do better in PD if your opponent cooperates, a behavior that elicits cooperation is more effective than one that doesn't.
You might say It's effective at being affective.
For the real world anekdote, all successful businesses that i worked with seem to be tit-for-tat-ers (i never tried to screw anyone and never got screwed, and i'm pretty sure they do apply tit for tat, or else the defectors would be more common). There could be some sneaky cheaters out there, but mostly due to misplaced beliefs of shareholders that a sociopath CEO will do better than tit-for-tat CEO.
An apparent counter-example (Goldman Sachs): http://www.nytimes.com/2012/03/17/opinion/nocera-the-good-bad-and-ugly-of-capitalism.html?_r=2
I'm not contradicting your data. Just wanted to note that it's not a priori obvious that the iterated PD (which is a very simple problem specification) is a good approximation to real life competitive companies. Or at least that companies' problems factorize in a way that gives the PD as a factor.
We're used to human-human relation management (which the OP says is well modeled by PD), and so human CEOs apply that in their relationships with other human CEOs.
What do you think? Is the explanation as simple as you imply: that company success has such a strong factor of human-human relationships with people representing/heading other companies, that good management there (Tit for Tat) can swamp other management considerations that have no exact human relations analog (creating a good product, differentiation, pricing, investments, supply chain issues, etc).
Well, normally if you are dealing with a company as a partner, it's a company for which the partnerships are a huge factor - e.g. in my case re-distributors.
The iterated prisoner's dilemma is an oversimplified problem. In the real world, usually, the non-cooperative players are known for their non-cooperative behaviours try channels other than getting screwed over yourself.
Furthermore, there are disparity of force cases where everyone's defecting against other party, just because other side can not retaliate. E.g. you could of paid me $0.5 per day to do the work I normally do, if I had no computer of my own or ability to go work at other employer, and I would have to do the work while barely surviving - and you could then pocket well over 99% of the income. That's how it works with outsourcing to third world.